Spatiotemporal Complexity of Continental

Bulletin of the Seismological Society of America, Vol. 99, No. 1, pp. 52–60, February 2009, doi: 10.1785/0120080005

Spatiotemporal Complexity of Continental Intraplate Seismicity: Insights from Geodynamic Modeling and Implications for Seismic Hazard Estimation by Qingsong Li, Mian Liu, and Seth Stein

Abstract Continental intraplate seismicity seems often episodic, clustered, and migrating. The observed seismicity shows both spatial clustering in seismic zones and scattering across large plate interiors, temporal clustering followed by long periods of quiescence, and migration of seismicity from one seismic zone to another. Here, we explore the complex spatiotemporal patterns of intraplate seismicity using a 3D visco- elasto-plastic finite-element model. The model simulates tectonic loading, crustal failure in earthquakes, and coseismic and postseismic stress evolution. For a laterally homogeneous lithosphere with randomly prespecified perturbations of crustal strength, the model predicts various spatiotemporal patterns of seismicity at different timescales: spatial clustering in narrow belts and scattering across large regions over hundreds of years, connected seismic belts over thousands of years, and widely scattered seismicity over tens of thousands of years. The orientation of seismic belts coincides with the optimal failure directions associated with the assumed tectonic loading. Stress triggering and migration cause spatiotemporal clustering of earthquakes. When weak zones are included in the model the predicted seismicity initiates within the weak zones but then extends far beyond them. If a fault zone is weakened following a large earthquake, repeated large earthquakes can occur on the same fault zone even in the absence of strong tectonic loading. These complex spatiotemporal patterns of intraplate seismicity predicted in this simple model suggest that assessment of earthquake hazard based on the limited historic record may be biased toward overestimating the hazard in regions of recent large earthquakes and underestimating the hazard where seismicity has been low during the historic record.

Introduction The distribution of earthquakes in space and time within faults that remain active for some time and then have long continental interiors is far more complex than on major plate quiescent periods during which other fault zones are active boundary faults. Earthquakes on plate boundaries result from (Crone et al., 2003; Camelbeeck et al., 2007). Thus, some strain produced by steady relative plate motions. Hence, faults without historically recorded earthquakes, such as the plate boundaries remain the loci of large quasi-periodic Meers fault in Oklahoma (Crone and Luza, 1990), appear to earthquakes for long intervals until the plate boundary ge- have had prehistoric large events (Clark and McCue, 2003). ometry changes. As a result, the long-term pattern of seis- Besides individual faults, earthquakes in continental in- micity inferred from geological fault studies is generally traplate regions also appear to occur in temporal clusters. For consistent with that observed in large earthquakes in historic/ example, earthquakes in North China’s Weihe-Shanxi graben instrumental records (e.g., Marco et al., 1996; Weldon et al., are temporally clustered before 1700 (Fig. 2a). Similarly, 2004; Cisternas et al., 2005). Moreover, the direction and large earthquakes (M>6) in the North China Plain are clus- rate of geodetically observed strain accumulation is generally tered between 1900 and 2000 (Fig. 2b). Moreover, the locus consistent with the mechanisms and recurrence of large of seismicity seems to be migrating from the Weihe-Shanxi earthquakes (Stein and Freymueller, 2002). graben eastward to the North China Plain in the past 200 yr The situation is quite different within continental plate (Fig. 1b). interiors, where earthquakes appear to be clustered, episodic, The spatial patterns of continental intraplate seismicity and migrating in space (Fig. 1). Paleoseismic data show that are also complex, with both spatial clustering of earthquakes intraplate earthquakes often occur in temporal clusters on in seismic zones and scattered activity over large regions

52 Spatiotemporal Complexity of Continental Intraplate Seismicity 53

Figure 1 Intracontinental earthquakes in (a) the CEUS and (b) North China. Dashed lines in (a) show the rift zones: midcontinent rift, MCR; and Reelfoot rift, RR. Data for the CEUS are from the National Earthquake Information Center (NEIC) catalog. Earthquake data of North China are from China Earthquake Administration. Gray dots in CEUS represent earthquakes before 1973; white dots represent earthquakes after 1973. Gray dots in North China represent earthquakes before 1900; white dots represent earthquakes after 1900. Thick lines are regional trends of seismic belts; black arrows are the direction of maximum compressive stress.

(Fig. 1). For example, historic and instrumentally recorded earthquakes in the central and eastern United States (CEUS) appear to be concentrated in several zones, notably the New (a) 9 Madrid, Charleston, and Eastern Tennessee seismic zones Weihe-Shanxi Graben (Fig. 1a). In addition, scattered seismicity occurs outside 8 these zones. Some seismic zones, such as the New Madrid zone, occur along the failed rift zones, but other seismic zones do not (Schulte and Mooney, 2005). Conversely, some 7 prominent rifts, such as the midcontinent rift in central North America, have little seismicity (Fig. 1a). As a result, the

6 relation between continental intraplate seismicity and geologic structure remains uncertain and the subject of ac-

Earthquake Magnitude tive investigation, as summarized by papers in Stein and 5 Mazzotti (2007). 1300 1400 1500 1600 1700 1800 1900 2000 The complex variation of seismicity in space and time Time (year) poses a major challenge for seismic hazard estimation. It seems likely that inferring seismic hazards from historical (b) 9 seismicity can overestimate the hazard where recent earth- North China Plain quakes have occurred and underestimate the hazard else- 8 where (e.g., Swafford and Stein, 2007; van Lanen and Mooney, 2007). To reduce this difficulty, researchers have developed hazard maps that also reflect geologic structure, 7 yielding maps that predict more diffuse hazards (Halchuk and Adams, 1999; Tóth et al., 2006). 6 Numerous models have attempted to explain the com- plexities of intraplate seismicity. Generally, these models in-

Earthquake Magnitude volve temporal variations of the physical properties of faults 5 and the stress acting on them. Crone et al. (2003) suggest that 1300 1400 1500 1600 1700 1800 1900 2000 Time (year) temporal clustering, in which faults turn on to generate a series of large earthquakes and then turn off for a long time, Figure 2 Historic records of intraplate earthquakes (M>6)in may reflect the evolution of pore fluid pressure in the fault (a) the Weihe-Shanxi graben and (b) the North China Plain. zone (Sibson, 1992). In this model, low-permeability seals 54 Q. Li, M. Liu, and S. Stein

1 mm/yr form around the fault zone as stress accumulates, raising the pore pressure until an earthquake happens. Faults that weaken during earthquakes and heal during interseismic C phases could also contribute to the complexity of seismicity 1 mm/yr (Lyakhovsky et al., 2001). The migration of seismicity be- D tween faults may result from stress transfer, in which stress changes due to earthquakes affect the location of future B events (Stein, 1999; Chéry et al., 2001; Mueller et al., 2004; Rundle et al., 2006). Failure and viscous relaxation of the A lower crust may cause a sequence of fault ruptures in short recurrence intervals (Kenner and Segall, 2000). Most of these studies are focused on a single event or fault zone, and regional tectonic loading is often ignored (e.g., Kenner and Segall, 2000; Lyakhovsky et al., 2001; Crone et al., 2003). In this study we explore the complexity Figure 3 Finite-element model for case 1. The model is loaded of intraplate earthquakes and some of the controlling factors by compression at the eastern and western boundaries and by extension at the northern and southern boundaries. The light gray layer in a regional scale finite-element model. represents the brittle (elastoplastic) upper crust; the dark gray layer represents the ductile (viscoelastic) lower crust and upper mantle. A, B, C, and D are four small regions used for showing seismicity clustering in Figure 7. Finite-Element Model

Our model is built on the 3D visco-elasto-plastic finite- element model we developed for simulating lithosphere de- lower crust viscosity from postseismic deformation studies formation (Li and Liu, 2006, 2007). The model simulates (1019 Pa sec or less, e.g., Pollitz et al., 2001a) and the vis- tectonic loading, crustal failure in earthquakes, and the asso- cosity from postglacial rebound studies (∼1021 Pa sec, e.g., ciated coseismic and postseismic stress evolution. Tectonic Hager, 1991). The yield strength of the upper layer is de- loading is simulated by applying velocity boundary condi- scribed by a value of 50 MPa for the cohesion and 0.4 tions at the model boundaries. Crustal failure in an earth- for the coefficient of internal friction. These values are within quake is simulated by instantly decreasing strength in an the reasonable ranges for the upper crust and were used in element by a given amount when stress in the element previous modeling studies (e.g., Kenner and Segall, 2000). – – reaches the Drucker Prager failure criterion. The Drucker When crust fails, the stress drops by 1 MPa. The energy re- Prager criterion is similar to the Mohr–Coulomb criterion but lease is equivalent to an M ∼6 earthquake for one failed ele- is more numerically stable. Stress drop at the failed element ment (50 × 50 × 10 km). changes the stress in the surrounding elements. Further stress To minimize the impact of the poorly constrained initial changes may derive from postseismic viscous relaxation of stress state, we begin the numerical experiments after the the lower crust and the upper mantle. The coseismic and model has reached a quasi-steady state. Figure 4 shows an postseismic stress changes in the elements that failed may increase (or decrease) stress in the neighboring regions, thus example of how the predicted stress and earthquakes evolve. promoting (or inhibiting) earthquake occurrence there. Figure 4a shows the stress at the beginning of a 1000 yr pe- The model dimension is 2000 by 2000 km and is 60 km riod to be examined; the black dots are the earthquakes that thick (Fig. 3). A 20 km thick elastoplastic upper layer simu- occurred during this period. Figure 4b shows the stress at the lates the brittle upper crust (the schizosphere) and a 40 km end of this 1000 yr period. The earthquakes occur within re- thick viscoelastic lower layer simulates the ductile lower gions of high stresses. Each earthquake causes stress to drop crust and upper mantle (the plastosphere). Using different below the crustal yield strength in the failed elements. The thickness of these layers within reasonable ranges does not stress drop also perturbs stress in the neighboring. Hence, affect the main model results. The eastern and western the stress field in the model domain continuously evolves boundaries are compressed uniformly at 1 mm=yr, while with time. the northern and southern boundaries are extended at the same rate. The resulting strain rate is ∼5:0 × 1010 yr1, typ- ical for the deformation rate in continental intraplate regions Model Results (e.g., Newman et al., 1999; Gan and Prescott, 2001; Calais et al., 2005). The bottom boundary is free slip. Young’s mod- To illustrate the effect of various factors on the pre- ulus is 8:75 × 1010 Pa, and Poison’s ratio is 0.25 for both dicted spatiotemporal patterns of intraplate seismicity, we the crust and the mantle (Turcotte and Schubert, 1982). start with a simple model with a horizontally homogeneous For most cases we take the viscosity for the lower layer to crust and mantle and then incrementally increase the model be 1 × 1020 Pa sec, which is within the range between the complexity. Spatiotemporal Complexity of Continental Intraplate Seismicity 55

(a) (b) 500 km

0 0 650 km Proximity to strength (MPa)

-0.7 0

Figure 4 Evolution of stress and seismicity over a 1000 yr period in the model. (a) Initial stress level: the gray scale indicates how close the stress is to the crustal yield strength. The dots are epicenters of predicted earthquakes in the following 1000 yr. (b) The stress level at the end of the 1000 yr period.

Case 1: Timescale-Dependent Spatiotemporal ure events. To some extent, this perturbation simulates the Patterns of Seismicity lateral heterogeneity of the continental crust. The resulting seismicity shows both spatial clustering In this case, a randomly prespecified perturbation of the (in spots and belts) and scattering (over large regions) over crustal strength (Fig. 5), which is kept constant through the a period of 300 yr (Fig. 6a). Over a longer period, 3000 yr, modeling, is applied to an otherwise laterally homogeneous the earthquake clusters connect to form a network of seismic crust and mantle (Fig. 3). The magnitude of the perturbation zones (Fig. 6b). Over an even longer time, 30,000 yr, seis- 0:5 ranges from to 0.5 MPa, which are small values com- micity covers the entire model domain (Fig. 6c). This result pared with the crustal strength and stress drop in crustal fail- is expected because the model domain is horizontally homogeneous. Hence, seismicity is uniform over the long term but is concentrated in seismic zones during shorter intervals. 2000 The model results also show temporal clustering of km earthquakes in seismic zones. Figure 7 shows the predicted earthquake sequences for four 100 km2 regions, each con- taining four finite-element cells. Temporal clustering of earthquakes occurs in region B between 10,000 and 20,000 yr and in region C around 5000 and 22,000 yr. The temporal clustering is caused by coseismic and postseismic stress triggering among the neighboring elements. With- out coseismic and postseismic stress interaction between regions, we would expect periodic earthquake occurrence because the model has constant loading and constant stress release in earthquakes. The orientations of the predicted seismic belts, northwest–southeast or northeast–southwest, coincide with the optimal failure directions (maximum shear stress) resulting from the boundary conditions of east–west compression and north–south extension. 0 0 2000 km Strength Perturbation (MPa) Case 2: Effects of Weak Zones In this case we replace the random perturbation of the -0.5 0.5 initial crustal strength with three weak zones whose yield Figure 5 The random crustal strength perturbation used in strength is 5 MPa lower than the rest of the model domain case 1. (Fig. 8a). Two of the zones are parallel to each other and 56 Q. Li, M. Liu, and S. Stein

0 0 2000 km Time (yr) Time (yr) Time (yr)

0 300 0 3000 0 30,000

Figure 6 Seismicity in case 1 that occurred over timescales of (a) 300 yr, (b) 3000 yr, and (c) 30,000 yr. The gray scale of dots represents the time within each of the three time intervals. are perpendicular to the third. The initial stress is set to The spatiotemporal pattern of seismicity on different zero, and the model is constantly loaded at the boundaries. timescales is shown in Figure 8. On a short timescale Initially earthquakes occur only within the weak zones. After (600 yr), earthquakes occur both inside and outside the weak ∼80; 000 model years, stress in the surrounding crust be- zones (Fig. 8a). The seismicity shows some spatial clustering gins to reach the strength and cause failure. After another with no clear correlation with the weak zones. Over a longer 100,000 yr, model seismicity reaches a quasi-steady state, period (6000 yr), the seismicity forms belts extending from in which the predicted seismicity during one long period the weak zone (Fig. 8b). This pattern remains the same over (> 60; 000 yr) is similar to another. longer periods (Fig. 8c). Hence, on the long timescales, seis-

Crustal Failure (Region A) Crustal Failure (Region B) 3 3

2 2

1 1 Number of Failures Number of Failures

0 0 0 10000 20000 30000 0 10000 2000 30000 Time (yr) Time (yr)

Crustal Failure (Region C) Crustal Failure (Region D) 3 3

2 2

1 1 Number of Failures Number of Failures

0 0 0 10000 20000 30000 0 10000 20000 30000 Time (yr) Time (yr)

Figure 7 Temporal distribution of predicted earthquakes in regions A, B, C, and D shown in Figure 3. Spatiotemporal Complexity of Continental Intraplate Seismicity 57

Z3 Z1

0 0 2000 km Time (yr) Time (yr) Time (yr)

0 600 0 6000 0 60,000

Figure 8 Seismicity in case 2 that occurred over timescales of (a) 600 yr, (b) 6000 yr, and (c) 60,000 yr. The gray scale of dots represents the time within each of the three time intervals. The gray belts in (a) are prespecified weak zones. micity is not confined to the weak zones. Moreover, as in the a weak fault zone (zone Z1 in Fig. 8a), whose strength is previous case, the short-term seismicity pattern does not fully 5 MPa lower than that of the rest of the model domain. represent that over longer terms. We consider three different scenarios of fault weakening following a large fault rupture. In the first scenario the fault zone is weakened instantly Case 3: Effects of Fault Weakening by 0.5 MPa immediately after the initial rupture, which has a In the first two cases spatial variations in crustal strength stress drop of 1 MPa. This is the averaged stress drop in the cause the spatiotemporal variations in seismicity. In the third fault zone of finite width (∼70 km) and, thus, is lower than case we explore the effects of temporal variation in fault actual stress drops on the fault plane. The energy release of properties. In this case the crust has a uniform strength ex- this rupture event is equivalent to an M ∼8 earthquake. Stress cept along a single fault. This analysis is motivated by the on the fault zone then gradually rises due to the far-field load- fact that paleoearthquake and fault offset studies have found ing and viscous relaxation in the lower crust and the upper evidence of repeated ruptures on individual intraplate faults mantle (Fig. 9). When the stress in the fault zone reaches the and fault zones in short periods followed by long periods of new, lower fault strength in 1800 yr, a new rupture occurs quiescence (Tuttle et al., 2002; Crone et al., 2003). The short time between the repeated earthquakes on these intraplate Stress evolution in the weak zone faults, such as the ∼500 yr found for the recent large events in the New Madrid zone (Tuttle et al., 2002), is at odds with 44 the low rates of deformation and far-field loading in plate (a) interiors (Newman et al., 1999; Calais et al., 2005). Li et al. 42 (2005) have shown that a large intraplate earthquake can re- (c) duce stress in the fault zone such that thousands of years (b) would be needed for stress to rebuild for another similar rup- 40 ture. Thus, large intraplate earthquakes repeating in short Stress (MPa) time intervals would require either some kind of local load- 38 ing to build up the stress or local weakening to permit repeated failure at lowered stress levels. A number of mechanisms for local loading and/or fault 36 weakening have been proposed (e.g., Kenner and Segall, -1000 0 1000 2000 3000 4000 5000 6000 2000; Pollitz et al., 2001b), although there is no direct evi- Time (yr) dence for the proposed weakening (McKenna et al., 2007). Figure 9 Stress evolution in the weak zone with different weak- Assuming some kind of fault weakening occurs in nature ening scenarios: (a) instant weakening, (b) continuous weakening, (Sibson, 1992; Lyakhovsky et al., 2001), we explore here and (c) continuous weakening that stops after 2000 yr. Gray lines its effect on seismicity. In this case we explicitly incorporate indicate the yield strength in each weakening scenario. 58 Q. Li, M. Liu, and S. Stein

(Fig. 9). Assuming no further fault zone weakening, it then nent loaded mainly at its boundaries, as represented in our takes another 3000 yr for the fault zone to rupture again case 1, the seismicity is expected to align with the maximum (Fig. 10a). The recurrence interval will be about one order shear stresses resulting from the loading conditions. In the of magnitude longer if the viscosity of lower crust and upper simplest case the maximum compressive stress bisects con- 21 mantle is one order of magnitude higher (> 10 Pa sec). jugate trends of seismicity. However, this does not seem the In the second scenario, the fault zone continuously case in the continents. The seismic belts in North China form weakens at a rate of 1 MPa per 1000 yr after the first rupture. conjugate north-northeast and southeast-east pairs (Fig. 1b), The stress in the weak zone recovers and is then released whereas the direction of present-day maximum compressive (Fig. 9) in a series of repeated earthquakes (Fig. 10b). tectonic stress is around N55°E (Yuan et al., 1999). The seis- The recurrence interval for the successive events is hundreds mic belts in CEUS are oriented northeast (Fig. 1a), close to of years. Increasing the viscosity of lower crust and upper the average maximum compressive direction of tectonic mantle increases little the recurrence interval because the stress of around N60°E (Zoback and Zoback, 1989). Thus, recurrence interval is largely controlled by the speed of the orientation of seismic zones may instead be controlled weakening. by the inherited structures, such as ancient fault zones, which In the third scenario, the fault zone gradually weakens in formed in a different stress field from the present one (John- the first 2000 yr following the initial rupture. After that the ston and Schweig, 1996). fault zone maintains a constant strength. The stress evolution When significant weak zones are present in the crust, the and rupture sequences in the first 2000 yr are the same as model predicts that intraplate seismicity is not confined to those in the second scenario. After the weakening stops, it these zones. Earthquakes occur all over, because the stress takes a longer time for ruptures to repeat (Fig. 9). The recurrence interval is even longer if the viscosity of lower crust everywhere in the model domain is near its critical value for and upper mantle is higher. Eventually, the recurrence inter- fault sliding. This condition is likely true for most continents val becomes long enough to be compatible with the rate of where seismicity is broadly diffuse. Nonetheless, our results far-field boundary loading. indicate that seismicity would concentrate along belts extending from the weak zones, oriented in the same directions Discussion as the weak zone and optimal failure directions. The regional temporal clustering in the model (Fig. 7) is The spatiotemporal patterns of seismicity predicted by caused by coseismic and postseismic stress triggering. Such our simple model have some similarities to those observed stress triggering may have caused a series of large earth- in continental interiors. For a laterally homogeneous conti- quakes to occur on nearby faults within a short period, as

(a) Rupture series (instant weakening) (b) Rupture series (continuous weakening) Rupture Rupture

-1000 0 1000 2000 3000 4000 5000 6000 -1000 0 1000 2000 3000 4000 5000 6000 Time (yr) Time (yr)

-1000 0 1000 2000 3000 4000 5000 6000 Time (yr)

Figure 10 Predicted crustal rupture sequences for (a) instant weakening, (b) continuous weakening, and (c) weakening that stops after 2000 yr. Spatiotemporal Complexity of Continental Intraplate Seismicity 59 in the 1811–1812 New Madrid seismic zone earthquakes 3. Fault weakening can lead to repeated earthquakes on (Mueller et al., 2004) and the 2005–2007 Sumatra earth- intraplate fault zones. The predicted temporal patterns quakes (McCloskey et al., 2005). However, our model of these repeated earthquakes vary with the weakening cases 1 and 2, in which constant fault properties and tectonic history. loading are assumed, cannot explain repeated earthquakes on Our model replicates some of the spatiotemporal com- individual faults in short periods, such as the approximately plexity of clustered, episodic, and migrating intraplate earth- 500 yr recurrence at the New Madrid seismic zone. Such re- quakes. Many factors not included in the model, such as a peated earthquakes on an individual fault may be predicted local driving force or that from the base of the lithosphere, by the model if fault weakening is assumed (Fig. 10). and variable viscous rheology in the lower crust and upper Although there is no direct evidence for fault weak- mantle, may further contribute to the complexity of continen- ening, several possible causes have been proposed. Sibson tal intraplate earthquakes. Our results of timescale-dependent (1992) suggested fault weakening through change of water spatiotemporal patterns of intraplate seismicity support the pore pressure on faults. Lyakhovsky et al. (2001) suggested suggestions that seismicity patterns observed from short- changes of fault rheology due to a balance between weaken- term seismic records may not reflect the long-term patterns ing by earthquakes and interseismic healing. Fault weaken- of intraplate seismicity (e.g., Crone et al., 2003; Camelbeeck ing and healing may help to explain the paleoseismic data et al., 2007). Because of the limited earthquake records in that have been interpreted to indicate for a series of large most places, assessment of earthquake hazard may be biased earthquakes in the New Madrid seismic zone with recurrence toward overestimating the hazard in regions of recent large intervals of a few hundred years (Schweig and Ellis, 1994; earthquakes and underestimating the hazard where seismic- Holbrook et al., 2006). Kenner and Segall (2000) tried to ity has been quiescent. simulate this inferred seismic sequence in a viscoelastic model. By assuming a sudden weakening of a localized weak lower crust embedded in an elastic crust, their model shows Data and Resources that weakening in the lower crust can causes stress amplifi- cation in the upper crust, resulting in a sequence of ruptures Seismicity data for the CEUS are from the National of the fault zone. Their predicted interseismic strain rate is Earthquake Information Center (NEIC) catalog (http://neic below 1 × 107 yr1, below the detection level of previous .usgs.gov/neis/epic, last accessed June 2007). Earthquake Global Positioning System (GPS) surveys in the New Madrid data for North China are from China Earthquake Adminis- seismic zone. Hence, the new zero crustal motion is indicated tration; these data may be obtained by contacting research by the GPS data (Calais et al., 2005; Smalley et al., 2005). scientist in the China Earthquake Administration. However, the cause of the sudden lower crust weakening is left for speculation. Alternatively, we have shown that the Acknowledgments clusters of large intraplate earthquakes can result from fault weakening and healing, and the clusters can be separated by This work is partially supported by U.S. Geological Survey (USGS) Grant Number 04HQGR0046, National Science Foundation (NSF) Division long periods of quiescence. of Earth Sciences (EAR) Grant Number 0225546, and NASA Cooperative Agreement Number NCC5-679. We thank Editor Fred Pollitz, Susan Hough, and an anonymous reviewer for their constructive comments. Conclusions We have explored the spatiotemporal patterns of intra- References plate earthquakes with a 3D finite-element model. The model Calais, E., G. Mattioli, C. Demets, J. M. Nocquet, S. Stein, A. Newman, and includes far-field boundary loading, crustal failure in earth- P. Rydelek (2005). Tectonic strain in plate interiors?, Nature 438, quakes, coseismic and postseismic stress evolution, and fault E9–E10. weakening. Our main results include the following: Camelbeeck, T., K. Vanneste, P. Alexandre, K. Verbeeck, T. Petermans, P. Rosset, M. Everaerts, R. Warnant, and M. Van Camp (2007). Rele- 1. Even with a horizontally homogeneous crust and mantle vance of active faulting and seismicity studies to assess long term the model predicts complex spatiotemporal seismicity earthquake activity in Northwest Europe, in Continental Intraplate patterns that vary with timescales. Over hundreds of years Earthquakes: Science, Hazard, and Policy Issues, S. Stein and S. Maz- the seismicity may appear both clustered in small belts zotti (Editors), Special Paper 425, Geol. Soc. Am., Boulder, Colorado, 193–224. and scattered over a larger region. Over thousands of Chéry, J., S. Merkel, and S. Bouissou (2001). A physical basis for time clus- years, the seismic clusters (belts) connect to form a net- tering of large earthquakes, Bull. Seismol. Soc. Am. 91, 1685–1693. work of seismic belts. Over tens of thousands of years, Cisternas, M., B. F. Atwater, F. Torrejón, Y. Sawai, G. Machuca, M. Lagos, scattered seismicity covers the whole model region. A. Eipert, C. Youlton, I. Salgado, T. Kamataki, M. Shishikura, C. P. 2. Coseismic and posterseismic triggering can cause tem- Rajendran, J. K. Malik, Y. Rizal, and M. Husni (2005). Predecessors of the giant 1960 Chile earthquake, Nature 437, 404–407. poral clustering of regional seismicity. If weak zones are Clark, D., and K. McCue (2003). Australian paleoseismology: towards incorporated in the model, the seismicity initiates in, but a better basis for seismic hazard estimation, Ann. Geophys. 46, is not confined to, these zones. 1087–1105. 60 Q. Li, M. Liu, and S. Stein

Crone, A. J., and K. V. Luza (1990). Style and timing of Holocene surface Schweig, E. S., and M. A. Ellis (1994). Reconciling short recurrence inter- faulting on the Meers fault, southwestern Oklahoma, GSA Bull. 102, vals with minor deformation in the New Madrid seismic zone, Science 1–17. 264, 1308–1311. Crone, A. J., P. M. De Martini, M. N. Machette, K. Okumura, and J. R. Sibson, R. H. (1992). Implications of fault-valve behaviour for rupture nu- Prescott (2003). Paleoseismicity of two historically quiescent faults cleation and recurrence, Tectonophysics 211, 283–293. in Australia: implications for fault behavior in stable continental re- Smalley, R., M. A. Ellis, J. Paul, and R. B. Van Arsdale (2005). Space geo- gions, Bull. Seismol. Soc. Am. 93, 1913–1934. detic evidence for rapid strain rates in the New Madrid seismic zone of Gan, W., and W. H. Prescott (2001). Crustal deformation rates in central and central USA, Nature 435, 1088–1090. eastern U.S. inferred from GPS, Geophys. Res. Lett. 28, 3733–3736. Stein, R. S. (1999). The role of stress transfer in earthquake occurrence, Na- Hager, B. H. (Editor) (1991). Mantle Viscosity: A Comparison of Models ture 402, 605–609. from Postglacial Rebound and from the Geoid, Plate Driving Forces, Stein, S. and J. T. Freymueller (Editors) (2002). Plate Boundary Zones: Con- and Advected Heat Flux, Kluwer Academic Publishers, London. cepts and Approaches, American Geophysical Union, Washington, Halchuk, S., and J. Adams (1999). Crossing the border: assessing the dif- D.C., 425 pp. ferences between new Canadian and American seismic hazard maps, in Stein, S. and S. Mazzotti (Editors) (2007). Continental Intraplate Earth- Proceedings Eighth Canadian Conference on Earthquake Engineer- quakes: Science, Hazard, and Policy Issues, Special Paper 425, Geol. ing,77–82. Soc. Am., Boulder, Colorado. Holbrook, J., W. J. Autin, T. M. Rittenour, S. Marshak, and R. J. Goble Swafford, L., and S. Stein (2007). Limitations of the short earthquake record (2006). Stratigraphic evidence for millennial-scale temporal clustering for seismicity and seismic hazard studies, in Continental Intraplate of earthquakes on a continental-interior fault: holocene Mississippi Earthquakes: Science, Hazard, and Policy Issues, S. Stein and S. Maz- River floodplain deposits, New Madrid seismic zone, USA, Tectono- zotti (Editors), Special Paper 425, Geol. Soc. Am., Boulder, Colorado, physics 420, 431–454. 49–58. Johnston, A. C., and E. S. Schweig (1996). The enigma of the New Madrid Tóth, L., E. Győri, P. Mónus, and T. Zsíros (2006). Seismic hazard in the earthquakes of 1811–1812, Annu. Rev. Earth Planet. Sci. 24, 339–384. Pannonian region, in The Adria Microplate: GPS Geodesy, Tectonics Kenner, S. J., and P. Segall (2000). A mechanical model for intraplate earth- and Hazards, N. Pinter, G. Gyula, J. Weber, S. Stein and D. Medak quakes; application to the New Madrid seismic zone, Science 289, (Editors), Springer, Netherlands, 369–384. 2329–2332. Turcotte, D. L., and G. Schubert (1982). Geodynamics: Applications of Con- Li, Q., and M. Liu (2006). Geometrical impact of the San Andreas fault tinuum Physics to Geological Problems, John Wiley & Sons, New on stress and seismicity in California, Geophys. Res. Lett. 33, L08302, York, 369–384. doi 08310.01029/02005GL025661. Tuttle, M. P., E. S. Schweig, J. D. Sims, R. H. Lafferty, L. W. Wolf, and M. L. Li, Q., and M. Liu (2007). Initiation of the San Jacinto fault and its inter- Haynes (2002). The earthquake potential of the New Madrid seismic action with the San Andreas fault: insights from geodynamic model- zone, Bull. Seismol. Soc. Am. 92, 2080–2089. ing, Pure Appl. Geophys. 164, 1937–1945. van Lanen, X., and W. D. Mooney (2007). Integrated geologic and geo- Li, Q., M. Liu, and E. Sandvol (2005). Stress evolution following the 1811– physical studies of North American continental intraplate seismicity, 1812 large earthquakes in the New Madrid seismic zone, Geophys. in Continental Intraplate Earthquakes: Science, Hazard, and Policy Res. Lett 32, L11310, doi 10.1029/2004GL022133. Issues, S. Stein and S. Mazzotti (Editors), Special Paper 425, Geol. Lyakhovsky, V., Y. Ben-Zion, and A. Agnon (2001). Earthquake cycle, fault Soc. Am., Boulder, Colorado, 101–112. zones, and seismicity patterns in a rheologically layered lithosphere, J. Weldon, R., K. Scharer, T. Fumal, and G. Biasi (2004). Wrightwood and the Geophys. Res. 106, 4103–4120. earthquake cycle: what a long recurrence record tells us about how Marco, S., M. Stein, A. Agnon, and H. Ron (1996). Long term earthquake faults work, GSA Today 14, 4–10. clustering: a 50,000 year paleoseismic record in the Dead Sea graben, Yuan, J., J. Xu, and S. Gao (1999). Inversion analysis of present tectonic J. Geophys. Res. 101, 6179–6192. stress field in North China region based on GPS data, Geol. Sci. Tech. McCloskey, J., S. S. Nalbant, and S. Steacy (2005). Indonesian earthquake: Inf. 18, 99–103. Earthquake risk from co-seismic stress, Nature 434, 291–291. Zoback, M. L., and M. D. Zoback (1989). Tectonic stress field of the con- McKenna, J., S. Stein, and C. Stein (2007). Is the New Madrid seismic zone tinental United States, in Geophysical Framework of the Continental hotter and weaker than its surroundings?, in Continental Intraplate United States, L. C. Pakiser and W. D. Mooney (Editors), Geol. Soc. Earthquakes: Science, Hazard, and Policy Issues, S. Stein and S. Maz- Am. Memoir, Vol. 172, 523–539. zotti (Editors), Special Paper 425, Geol. Soc. Am., Boulder, Colorado, 167–175. Mueller, K., S. E. Hough, and R. Bilham (2004). Analysing the 1811–1812 Lunar and Planetary Institute New Madrid earthquakes with recent instrumentally recorded after- Universities Space Research Association shocks, Nature 429, 284–288. 3600 Bay Area Boulevard Houston, Texas 77058 Newman, A., S. Stein, J. Weber, J. Engeln, A. Mao, and T. Dixon (1999). [email protected] Slow deformation and lower seismic hazard at the New Madrid seismic (Q.L.) zone, Science 284, 619–621. Pollitz, F. F., C. Wicks, and W. Thatcher (2001a). Mantle flow beneath a continental strike-slip fault; postseismic deformation after the 1999 Department of Geological Sciences Hector Mine earthquake, Science 293, 1814–1818. University of Missouri Pollitz, F. F., L. Kellogg, and R. Bürgmann (2001b). Sinking mafic body in a Columbia, Missouri 65211 reactivated lower crust: a mechanism for stress concentration at the (M.L.) New Madrid seismic zone, Bull. Seismol. Soc. Am. 91, 1882–1897. Rundle, J. B., P. B. Rundle, A. Donnellan, P. Li, W. Klein, G. Morein, D. L. Department of Earth and Planetary Sciences Turcotte, and L. Grant (2006). Stress transfer in earthquakes, hazard Northwestern University estimation and ensemble forecasting: inferences from numerical simu- – Evanston, Illinois 60208 lations, Tectonophysics 413, 109 125. (S.S.) Schulte, S. M., and W. D. Mooney (2005). An updated global earthquake catalogue for stable continental regions: reassessing the correlation with ancient rifts, Geophys. J. Int. 161, 707–721. Manuscript received 1 April 2008